What’s on Your Mind?

Brain-computer interfaces (BCIs) have been around in one form or another for several decades. The earliest attempts to build BCIs were more in the realm of experimentation than practicality, but with some recent notable successes at companies such as Neuralink, it looks like the technology might finally be starting to come of age. These systems could ultimately transform all of our lives, but initially the biggest impact is expected to be felt by those with medical conditions impacting their ability to move or speak normally.

In these cases, an effective BCI could give its user the ability to interact with computers and communicate with the people around them. In order to do that, the system must first learn to decode the thoughts of its user. This is a tricky business, however a team led by researchers at Stanford University has just reported on some significant progress in the area. They have developed an artificial intelligence (AI)-based algorithm that can analyze brain activity measurements collected by a BCI and use it to decode a person’s inner thoughts.

So technically, yes, they did build a mind reader. But there are some caveats, so don’t put on your tin foil hat just yet. First of all, to pick up much of anything at all, the subject must intentionally focus on the word or words that they want the system to recognize. And more importantly yet, that subject must have electrodes physically implanted in their brain. So unless you recently woke up with a large, inexplicable scar on your head along with a strange port that you don’t recall ever having been there, you should be just fine.

To build the system, the researchers recorded neural activity from microelectrodes implanted in the motor cortex of the brain in four individuals with severe paralysis. Each of these individuals was asked to attempt to speak, or imagine speaking, a set of words during the recording sessions. The imagined speech data was then utilized to train an AI algorithm to interpret its meaning. When tested, it was found that this algorithm could successfully decode inner thoughts from a 125,000-word vocabulary with as high as a 74% accuracy rate.

Recognizing that many people will be concerned about the implications of their work, the team devised a scheme in which the inner thoughts of a user cannot be decoded unless they first think of a password to explicitly allow it. You are probably thinking this sounds about as safe as a bank vault secured with a sticky note that says "no thieves allowed," and you are right. But we cannot shun any technology that has the potential to be abused or we will find ourselves back in the Stone Age. The potential benefits are massive and must be pursued, but as these systems improve, more robust privacy protections will need to be devised.